Dynamic Resource Management In Clouds With Support For Federation

In this paper, we addressed the problem of multi-objectiveoptimization of resource management in cloud federation. The objectivespartly come from cloud clients specifying in Service Level Agreement,including performance, availability, cost and other requirements. Whileother objectives come from cloud providers, who desire to save operatingcost by reducing energy consumption in order to achieve high profit. As animportant role of cloud computing, carrying out “pay-as-you-go”, cloudproviders always make every effort to use lowest resources to meet allrequirements from clients. Consequently, the resource management incloud should balance the benefits of different shareholders. It is obviouslythat more resources ensure the better performance and availability, butexcess resources not only waste usage costs of clients but also aggrandizeoperating costs of providers. On the other hand, insufficient resources canreduce operating costs of providers, but the requirements of clients maynot be satisfied, which brings punishment of violation to providers. As aresult, designing a multi-objective optimization of resource managementhas become a significant concern both for cloud clients and providers.With the development of virtualization, more and more enterpriseshave built their own private cloud for managing heterogeneous physicalservers in order to reduce operating costs. However, due to the limitationof resources, private cloud cannot satisfy the increasing requirement ofapplications. It is a trend that private cloud providers will rent someresources from public cloud providers to pull through. Cloud federationconstitutes. Under the environment of cloud federation, the resource management faces more austere challenges.We proposed a multi-objective optimization resource managementapproach in cloud federation. We analyzed the restrictive relation ofobjectives came from cloud clients and providers. The resourcemanagement problem was modeled as a multi-objective problem andsolved by serialized optimization with priorities, in which performanceowns the highest priority and then cost, availability and the resourceutilization. Meanwhile, cloud federation was divided to three types,vertical, horizontal and hybrid, according to the layered characteristic ofweb applications. For purpose of supporting federation, the cloudfederation center was designed, taking charge of registering, matching andcoordinating cloud services, managing cloud providers andconstructing/deconstructing cloud federation.Finally, to verify the effectiveness of our approaches, themulti-objective optimization of resource management in cloud federation,we designed and implemented two series of experiments by simulating twoclouds. We used two sets of real workloads to evaluate the whole approachincluding forecasting workload, calculating the amount of virtual machines,improving availability by virtual machine placement, saving cost by virtualmachine consolidation. The results of two experiments showed that theapproach we presented can achieve the objectives for sakes of cloud clientsand providers.In summary, we addressed the new challenges raised bymulti-objective resource management for cloud federation and designed anovel multi-objective optimization approach to solve this problem. In thisapproach every objectives was achieved sequentially as according to thepriority levels. Finally, the win-win situation of cloud clients and providerswere achieved.